PHOTO ILLUSTRATION: ANDREA D’AQUINO
“There is always a struggle over how technology is going to be used and who benefits from it,” says Institute Professor in the Department of Economics Daron Acemoglu. “And when technology goes down the path of automation, it’s not very beneficial for workers.” Heeding that lesson is crucial when considering the role AI will play in society. When it is used to automate tasks that can be performed by humans—as has often been the case up until this point—it will lead to worker replacement and lost jobs. But history has another lesson as well, he says: “An alternative path is often open, where you can use technology in a more pro-worker way.”
Acemoglu has spent much of his career looking at lessons of history. With University of Chicago Harris School of Public Policy political scientist James A. Robinson, he wrote the acclaimed Why Nations Fail: The Origins of Power, Prosperity and Poverty and its follow-up, The Narrow Corridor: States, Societies, and the Fate of Liberty, about how states can balance political liberty and economic success. More recently, he and Simon Johnson PhD ’89, the Ronald A. Kurtz Professor of Entrepreneurship at the MIT Sloan School of Management, wrote Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. Acemoglu, Johnson, and Ford Professor of Economics David Autor are faculty co-directors of the new MIT Shaping the Future of Work initiative, which will identify innovative ways to improve job quality and labor market opportunities for non-college workers. “We have a lot to learn from the last 1,000 years,” Acemoglu says.
Scientific advances distributed unequally
While technology has led to material success, it’s also led to rampant inequality. From industrial robots to computers, recent technologies have been often used to replace human workers, concentrating wealth in the hands of owners. Yet that has not always been the case. Railways, for example, fundamentally disrupted the means of production and transport in a way that ultimately led to the creation of more jobs, including train engineers, conductors, firemen, ticket sellers, and maintenance workers.
Some digital technologies have also created jobs, when governments and firms have invested in helping workers attain new skills, as in Germany and Japan, by training workers to do more maintenance and inspection tasks as robots were introduced.
We stand at the same crossroads with AI, says Acemoglu, who has researched the topic for more than a decade—long before the recent explosion of generative AI. Thus far, he says, “the benefits of automation have been hugely exaggerated, and the productivity benefits are not as major as people have expected.” In a 2022 paper for the Journal of Labor Economics titled “Artificial Intelligence and Jobs,” for example, he and several colleagues found companies that started posting more AI-related vacancies between 2010 and 2018 stopped asking for some skills and began demanding new capabilities, but reduced overall hiring, a sign that AI replaced human workers and offered relatively little productivity benefit.
“We found that the establishments hiring people with specialized AI skills were precisely those that have tasks that can be automated by AI, and that once they do that, they start cutting their hiring,” Acemoglu says. It doesn’t have to be that way, however, given the incredible versatility of AI. Almost by definition, AI has capabilities that are very distinct from human skills, he argues, doing things well that humans don’t—but also not being as good at some things humans do well. “That creates the possibility for human-machine symbiosis.”
For example, AI is very good at analyzing large amounts of data and diagnosing problems. In fields such as education and health care, that can lead to the development of personalized strategies for teaching or treating patients by augmenting rather than replacing the skills of teachers and doctors. “By identifying what challenges a specific student is facing, it’s a pathway to personalized education, which would otherwise be prohibitively expensive,” Acemoglu points out. The same can be said for manufacturing or specialized workers, such as electricians, currently in short supply. AI could be used as a training and diagnostic tool for troubleshooting problems, improving worker productivity without replacing workers.
A fresh look at AI
Really shifting from automating to augmenting human tasks will take a sea change in the way we consider AI, Acemoglu says. Government would need to create incentives to this kind of AI research through the National Institutes of Health and the National Science Foundation, subsidizing AI in the same way it helped get the renewable energy industry off the ground. Changes could be made in the tax code, which currently levies higher taxes on firms that hire human workers than those that implement digital replacements. Policies could change around data ownership, compensating knowledge producers for the data they generate that AI models are now appropriating for free.
Just as importantly, says Acemoglu, the norms of the AI industry need to change, from its current attitude of “move fast and break things” to a more contemplative and nuanced approach that examines the consequences of its actions on the workplace and society overall. “We are unleashing technologies that are going to affect millions of people, and it’s important to have a recognition that these tools will have major seen and unforeseen consequences,” he says. The good news is that it is not too late to shift course; the choices we make in the near future will determine whether AI shapes work and society for the worse or for the better. “There is time, but not oodles of time,” Acemoglu says. “We are starting from the position that the industry is very powerful and regulatory muscle is lacking. On the other hand, there has been a true transformation in the conversation on AI over the last few years, and a much greater agreement that workers need to be at the table.